symbols-printf
(Heavily inspired by Data.Symbol.Examples.Printf)
ghci> putStrLn $ printf @"You have %.2f dollars, %s" 3.62 "Luigi"
You have 3.62 dollars, Luigi
An extensible and type-safe printf from parsing GHC TypeLits Symbol literals,
matching the semantics of Text.Printf in base. The difference is that
the variants here will always fail to compile if given arguments of the wrong
type (or too many or too little arguments). Most of the variants also provide
useful type feedback, telling you the type of arguments it expects and how many
when queried with :t
or with typed holes.
There are three main calling conventions offered:
ghci> putStrLn $ printf @"You have %.2f dollars, %s" 3.62 "Luigi"
You have 3.62 dollars, Luigi
ghci> putStrLn $ pprintf @"You have %.2f dollars, %s" (PP 3.62) (PP "Luigi")
You have 3.62 dollars, Luigi
ghci> putStrLn $ rprintf @"You have %.2f dollars, %s" (3.62 :% "Luigi" :% RNil)
You have 3.62 dollars, Luigi
Comparing their types:
ghci> :t printf @"You have %.2f dollars, %s" 3.62 "Luigi"
FormatFun '[ .... ] fun => fun
ghci> :t pprintf @"You have %.2f dollars, %s" 3.62 "Luigi"
PP "f" -> PP "s" -> String
ghci> :t rprintf @"You have %.2f dollars, %s" 3.62 "Luigi"
FormatArgs '["f", "s"] -> String
-
The type of printf
doesn't tell you immediately what you
you need. However, if you do try to use it, the type errors will guide you
along the way, iteratively.
ghci> printf @"You have %.2f dollars, %s"
-- ERROR: Call to printf missing argument fulfilling "%.2f"
-- Either provide an argument or rewrite the format string to not expect
-- one.
ghci> printf @"You have %.2f dollars, %s" 3.62
-- ERROR: Call to printf missing argument fulfilling "%s"
-- Either provide an argument or rewrite the format string to not expect
-- one.
ghci> printf @"You have %.2f dollars, %s" 3.62 "Luigi"
You have 3.62 dollars, Luigi
ghci> printf @"You have %.2f dollars, %s" 3.62 "Luigi" 72
-- ERROR: An extra argument of type Integer was given to a call to printf
-- Either remove the argument, or rewrite the format string to include the
-- appropriate hole.
-
For pprintf
, it shows you need two arguments: A PP "f"
(which is a
value that supports being formatted by f
) like PP 3.62
, and a PP "s"
,
like PP "Luigi"
.
-
rprintf
tells you you need a two-item hlist (from "Data.Vinyl.Core"),
where the first item implements f
and the second item implements s
:
3.62 :% "Luigi" :% RNil
will do.
The following table summarizes the features and drawbacks of each
method:
Method |
True Polyarity |
Naked Arguments |
Type feedback |
printf |
Yes |
Yes |
Partial (via errors) |
pprintf |
Yes |
No (requires PP ) |
Yes |
rprintf |
No (HList-based) |
Yes |
Yes |
Ideally we would have a solution that has all three. However, as of now, we
have a "pick two" sort of situation. Suggestions are definitely welcome,
however, if you find something that satisfies all three benefits while still
allowing for polymorphism!
You can extend functionality with formatting for your own types by providing
instances of FormatType
.
Caveats
For medium-length or long strings, the parsing can be fairly slow and cause
slow compile times. This might be due to the underlying mechanism that the
symbols package exploits.
Moving to typechecker plugin based parsing does improve performance ...
however, I'm not sure how to get around requiring every module using printf
to require enabling the typechecker plugin, which isn't too great from a
usability standpoint. Template Haskell based alternatives (like
th-printf) already do require an extra pragma (for QuasiQuotes), though
so it might not be too bad in comparison.
Comparisons
There are a few other options for type-safe printfs out on hackage, and they
all differ in a few key ways. Some, like th-printf and
safe-printf, offer Template Haskell-based ways to generate your printf
functions. This package is intended as a "template-haskell free" alternative.
However, it is notable that with a Template-Haskell based approach, we can
solve the "pick two" situation above: th-printf's printf fulfills all
three requirements in the table above, with the only potential drawback being
Template Haskell usage.
Some others, like safe-printf, formatting, printf-safe,
xformat, and category-printf, require manually constructing your
fomatters, and so you always need to duplicate double-quotes for string
literals. This detracts from one of the main convenience aspects of printf,
in my opinion.
"You have " % f' 2 % " dollars, " % s
-- vs
"You have %.2f dollars, %s"
However, calling these libraries "safe printf libraries" does not do them
justice. A library like formatting is a feature-rich formatting library,
handling things like dates and other useful formatting features in a
first-class way that embraces Haskell idioms. This library here is merely a
type-safe printf, emulating the features of base's printf and C printf(3)
.
Todo
- Make faster
- Tests
- Support for localization/dynamic strings. Should be possible, but we'd
have to re-implement a subset of singletons.